Simultaneous Localization and Mapping (SLAM)

Core question in autonomous driving: how to represent the street map and self location with information from in-vehicle cameras and kinetic data (e.g. linear/angular velocity). Following are two Bayesian filters transforming visual/kinetic streams into latent variable series representing the map and location.

• Particle Filter: [code] [report]
• Unscented Kalman Filter: [code] [report]

Image Classification through Neural Networks

• A collection of convolutional neural network (CNN) models for classifying images into different categories. [code]

Seahorse Puzzle

• Nine square tiles to place in a 3-by-3 grid
• Around 100 billion possibilities with different positions and poses
• Only ONE possibility to match all patterns on the edges
• Solution: [code]

---

← Return to Main Page